High-precision cable force identification in bridges: a data-driven approach integrating digital twins and transfer learning

This paper proposes a novel data-driven framework for accurate cable force monitoring in bridges, which is crucial for ensuring structural integrity and extending service life. Leveraging the power of Digital Twins (DT), a comprehensive dataset of numerical cable models is generated. This synthetic data is used to pre-train a backpropagation (BP) neural network, providing a robust foundation. Subsequently, Transfer Learning (TL) is employed to refine the pre-trained model using limited real-world cable force data. This hybrid approach significantly enhances prediction accuracy, achieving consistently low errors—below 4% for short cables and 2% for long cables. Numerical examples demonstrate the effectiveness of the proposed framework, highlighting its potential for reliable and cost-effective cable force identification in real-world bridge applications.